{
  "version": 3,
  "created_at": "2026-07-07T16:33:37.678Z",
  "task_summary": "Phase 1 — exhaustive mining of public EoE omics datasets to surface dysregulated targets",
  "agents": [],
  "phases": [
    {
      "name": "Plan",
      "delegations": [
        {
          "steps": [
            {
              "title": "Build exhaustive dataset inventory",
              "description": "Programmatically enumerate every public EoE omics dataset across GEO (48 human series confirmed), ArrayExpress/BioStudies, PRIDE + MassIVE (proteomics), CELLxGENE + Broad Single Cell Portal (single-cell), and dbGaP/EGA (controlled-access genotype/WGS). For each, capture accession, assay type, platform, organism, tissue (esophageal biopsy/blood/plasma-EV), disease state (active/remission/control), age group (pediatric/adult), sample size per arm, treatment context, and access route. Deliverable: master curated inventory `eoe_dataset_inventory.csv` plus a summary figure `dataset_landscape.png` (datasets by assay type x sample size x year)."
            },
            {
              "title": "Tier and prioritize datasets for deep analysis",
              "description": "Rank inventory into tiers by relevance and tractability: Tier 1 = openly downloadable bulk RNA-seq/array with clear EoE-vs-control design and adequate n; Tier 2 = single-cell (GSE126250 T-cell atlas, epithelial scRNA-seq GSE218607/GSE168883); Tier 3 = complementary layers (methylation GSE186888, miRNA/ncRNA GSE275710/GSE36727, proteomics). Document inclusion/exclusion rationale. Deliverable: `dataset_tiers.csv` and a one-page prioritization rationale in the running report."
            },
            {
              "title": "Retrieve and QC Tier-1 bulk transcriptomics",
              "description": "Download processed matrices (and raw counts where available) for Tier-1 bulk RNA-seq and microarray series via GEOquery/direct supplementary files. Harmonize gene identifiers to HGNC/Ensembl, normalize (log-CPM/TMM for RNA-seq; RMA-consistent for arrays), and run QC: library-size/detection distributions, PCA colored by study/disease/age, outlier flagging. Deliverable: per-study QC panel `bulk_qc_panels.png`, harmonized expression checkpoints (`.parquet`), and a QC pass/fail log."
            },
            {
              "title": "Per-study differential expression",
              "description": "Run standardized EoE-vs-control differential expression per Tier-1 study (limma-voom for RNA-seq, limma for arrays), modeling available covariates (age group, sex, batch). Extract effect sizes and adjusted p-values. Cross-check recovery of the canonical EoE Diagnostic Panel genes (e.g., CCL26, POSTN, CLC, CPA3, DSG1, FLG, SPINK7). Deliverable: per-study DE tables (`de_<GSE>.csv`), volcano plots, and a concordance heatmap of top genes across studies (`de_concordance.png`)."
            },
            {
              "title": "Cross-study meta-analysis",
              "description": "Combine per-study effect sizes into a random-effects meta-analysis (e.g., metafor / Fisher or Stouffer combination) to derive a robust, reproducible EoE dysregulated-gene signature with per-gene direction, pooled effect, heterogeneity (I^2), and number of supporting studies. Stratify where powered (active vs remission; pediatric vs adult). Deliverable: `eoe_meta_signature.csv` (ranked) and figures `meta_forest_topgenes.png`, `meta_volcano.png`."
            },
            {
              "title": "Single-cell analysis of dysregulated populations",
              "description": "Process Tier-2 single-cell data (T-cell atlas GSE126250 + epithelial scRNA-seq) with scanpy: QC, integration/batch correction, clustering, cell-type annotation. Quantify disease-associated shifts in cell-type composition and run cell-type-resolved DE (EoE vs control), highlighting pathogenic Th2/effector populations, eosinophils, mast cells, and epithelial states. Deliverable: annotated UMAPs (`sc_umap_celltypes.png`, `sc_umap_disease.png`), composition bar/box plots, and per-cell-type DE tables."
            },
            {
              "title": "Pathway, regulator, and cell-cell communication analysis",
              "description": "On the meta-signature and cell-type DE, run pathway/GSEA enrichment (Reactome/KEGG/Hallmark), upstream-regulator inference (cytokine modules IL-13/IL-5/TSLP/IL-4), transcription-factor enrichment, and ligand-receptor / cell-cell communication (CellPhoneDB/LIANA on single-cell). Deliverable: enrichment dot plots (`pathway_enrichment.png`), regulator ranking table, and a ligand-receptor interaction figure (`ligand_receptor.png`)."
            },
            {
              "title": "Integrate complementary omics layers",
              "description": "Layer Tier-3 evidence onto the gene signature: DNA methylation (differential CpGs mapped to genes), miRNA/ncRNA (including plasma-EV biomarkers) with predicted targets, and available proteomics (differential proteins). Flag multi-omics-concordant genes (transcript + protein + regulatory support). Deliverable: `multiomics_integration.csv` and a concordance figure (`multiomics_concordance.png`)."
            },
            {
              "title": "Druggability and target-tractability annotation",
              "description": "Annotate the prioritized gene set with subcellular localization (secreted/plasma-membrane surfaceome relevant to protein-therapeutic design), Open Targets association scores, ChEMBL/known-drug status, tractability buckets, and STRING PPI neighborhood. Distinguish established EoE-pipeline targets from underexploited biology. Deliverable: `target_druggability.csv` and a PPI network figure (`target_ppi_network.png`)."
            },
            {
              "title": "Consolidate prioritized target shortlist and phase-1 report",
              "description": "Integrate all evidence into a weighted prioritization score (meta-effect strength + cross-study reproducibility + cell-type specificity + multi-omics concordance + tractability/novelty) to produce a ranked candidate target shortlist feeding phase 3 protein design. Write a structured markdown report synthesizing the data landscape, signature, cell populations, pathways, and shortlist with mechanistic rationale and caveats. Deliverable: `eoe_target_shortlist.csv` and `phase1_report.md` (with all figures embedded)."
            }
          ]
        }
      ],
      "id": "phase-0"
    }
  ],
  "feasibility": {
    "rationale": "Public EoE omics are abundant and directly retrievable: ~48 human GEO series (bulk RNA-seq, microarray, single-cell, methylation, miRNA/ncRNA), plus proteomics repositories and Open Targets/ChEMBL for druggability annotation. The main effort is harmonization and cross-study meta-analysis rather than data availability. Batch heterogeneity across studies and incomplete metadata are the principal risks, both manageable with standard covariate modeling and careful curation.",
    "confidence": "high"
  }
}